Team Scoring Systems and Swiss Chess Tournaments

Version 1.0
by Rob Brennan (brennanr@iworg.com <mailto:brennanr@iworg.com>) 17 Sept
'02
1. Introduction
Competition wargamers have long realised the advantages of the swiss
chess
method for running tournaments. This enthusiasm for the swiss chess
method
has not necessarily translated into a widespread appreciation of the
underlying assumptions of the method. It is a fact that the results
produced
by running swiss chess tournaments can easily be deeply flawed if
organisers
inadvertently violate the principals under which swiss chess was
designed to
be applied. Flawed is taken here to mean less likely to reward the
strongest player and more likely to be generate random results. Properly
applied, swiss chess is designed to be likely to produce a fair winner
ie
the same winner as would be found if a round-robin style of tournament
(where everyone plays everyone else) was played. An example of a
commonly
understood (if not always applied) principal of swiss chess is the fact
that
a certain minimum number of rounds are required to find the winner from
a
given number of players eg 16 players will generally require at least 4
rounds of play.
The choice of scoring system to be used is another aspect of the
competition
format that interacts with the swiss chess method. If competition
organisers
choose a scoring system without reference to the other parameters of the
tournament (number of players, number of rounds) then the results of
that
tournament can easily be just as flawed as if they played too few
rounds.
The effects of scoring system choice are perhaps more pernicious than
playing too few rounds because without understanding these effects the
competition wrongly appears to be equitably resolved. ie it looks like
you
have a fairly determined, clear-cut winner.
Within the DBM community there has long been debate on the relative
merits
of different scoring systems. There are essentially two camps based on
either the 0-10 scoring system presented in the rulesbook (or
derivatives
such as the BHGS 0-32 or GCS 0-15) or some form of Win/Draw/Loss scoring
(aka WDL, including 0-10 with +100VP for a win). It is not the purpose
of
this paper to focus on the details of individual game scoring, although
it
is a related issue (do you have a citation for where one might look to
address this issue?), but instead to focus on the issue of how best to
score
team results for team-based swiss chess competitions. This is seen as
especially important for two high profile DBM events, the Granson
International Team Challenge and the IWF World Team Championships. It is
hoped that this paper will explain why the current scoring arrangements
at
both of these events interact poorly with swiss chess and hence produce
unreliable (you should define this term or relate it back to flawed
which
you use above - unreliable seems a better term perhaps in that it more
nearly evokes the BLUE conditions for statistical estimators) outcomes.
Suggestions are made for scoring methods that would work better with
swiss
chess or other competition parameters that could be adjusted while
retaining
the current scoring methods to produce fairer outcomes. All of this work
is
presented with the intention of bringing this information to the
attention
of the organisers for everyone's benefit rather than any attempt to
undermine or diminish the excellent and much appreciated work of the
organisers in previous years.
Section 2 below outlines the vital relationship between scoring and
pairing
in swiss chess competitions. In section 3 you will find a discussion of
the
effects of the current scoring arrangements. Section 4 outlines some
proposals for possible alternate scoring systems and competition
mechanisms
to improve the fairness of outcomes. Finally section 5 presents some
conclusions.
2. Scoring and Pairing in Swiss Chess
Scoring in swiss chess competitions has one primary goal, to set up
pairings. Your score means nothing in absolute terms, it is the
people/teams
you play against that is important. This is because swiss chess is
trying to
estimate the best player/team by taking a limited sample of all possible
games ie when everyone plays everyone else. A simple illustration of
this
point is to consider what would happen if the top 10% of players/teams
in a
swiss chess competition played only against the bottom 10% of
players/teams - it is likely that all of the top players would receive a
maximum score.
The origins of using numerical vaules (scores) in swiss chess
competitions
to arrange pairings is just to have a simple way of recording how many
Wins
(1pt), Draws (0.5pts), and Losses (0pts) a given player has obtained.
This
is because the objective of swiss chess is to find the overall winner by
ensuring that they play a sample of the other strong players in the
comp. By
forcing the strong players to play each other swiss chess evaluates
their
relative strengths. Swiss chess assumes that the outcome of a game is a
reflection of the relative strengths of the two players involved.
Ideally (ie assuming no draws), swiss chess finds a winner by the
following
process for 16 players:
Round 1 generates 8 winners, Round 2 generates 4 double winners, Round 3
generates 2 triple winners, Round 4 generates 1 quadruple winner who is
the
winner of the competition. This person has won all their games and
played
all their games (after the first) versus other winners (who are also
assumed
to be strong players). The second part in italics is the important bit
as
that is what ensures that they have won their games against the other
strong
players in the field. This is ensured by making sure that the pairing is
correct. To translate this example into the WDL scoring which will
generate
pairings we see that after round 1 all the winners have 1 point, after
round
2 the double winners have 2 points etc. The fact that they are on the
same
score is then used to generate the required pairings. This is the ideal
situation without draws, when draws are added you in fact have to play
additional rounds to guarantee that the players will be seperated
sufficiently to determine a winner [1].
If we use a scoring system with more possible outcomes, eg 0-10 as
provided
in the DBM rulebook with 10-0, 9-1, 8-2, 7-3, 6-4 and 5-5 as the
possible
outcomes, then perhaps we are getting a more accurate estimate of the
relative skill levels of the players involved in a game. Does this help
our
swiss chess mechanism find a winner? No is the answer. The reason for
this
is that the additional possible game outcomes make our desired pairings
(between winners/strong players) less clear. If two players meet under
0-10
and they get a 7-3 result it tells us that player 1 (with the 7) is
stronger
but how does he compare to the player 1 of another game who got a 10-0?
It
could be that the second player 1 (with a 10) just played against a
weaker
player 2 and is in fact himself a weaker player than our first player 1.
This lack of information about the relative meaning of scores, ie both
scores weren't achieved against the same player 2, makes rewarding one
player 1 over another foolhardy. This is a critical difference in the
desired types of scoring for swiss chess competitions and round-robin
competitions where 0-10 would work fine. Do you mean 0-10 works fine for
round robin only? If so then perhaps a clearer wording might be
something
along the lines of: This is a critical difference in the desired types
of
scoring: For swiss chess competitions 0-10 does not work so well, while
for
round-robin competitions 0-10 would work fine
In practice, by having more possible scoring outcomes we make the
pairing of
strong players against each other harder to achieve. This is because the
groups of people on the same score have scoring information that is both
relative to the rest of the entire competition (how many wins they
posess)
and information relative to their exact opponents (how well they won
each
game). This results in a more continious distribution of score values
which
is more biased towards mean values. Thus all players tend to have a
score
which is nearer the average for the competition, thus pairings generated
from this information will tend to pair both weak and strong players
together throughout the competition. This violates the pairing
principals of
swiss chess. This means that it is easier to get a soft draw (because
everyone tends to be mixed up in the middle) and hence get the points to
lift you above everyone else for a competition win. Essentially this
means
that winning such a competition is largly determined by the luck of the
draw
rather than player skill. I wonder if this previous sentence wasnt a
bit
too strong. Although it will still be one of the stronger players who
tends
to win because you still have to win most of your games, it's just that
if
you win it is likely that someone else (who could be stronger than you)
received a much tougher draw than you. This problem can be solved by
playing
more rounds, just as we add rounds for the addition of draws to our
scoring
when using WDL [1].
It is also worth stating that scoring systems in swiss chess
competitions
have a secondary function and that is to seperate the placings of
players
after the final round when the idealised result of having only one
unbeaten
player is not achieved. Using the traditional WDL scoring it is assumed
that
players are ranked on the basis of most wins then most draws then least
losses. Thus a player with 3 wins and a draw places higher than a player
with 3 wins and a loss. This is not perfect because it does of course
take
into account the quality of the opposition faced and is thus biased
towards
the player with the softest draw. However soft draws are harder to
achive
under WDL scoring so the result is fairly robust. When two players on
the
same score, eg 3 wins and a draw (3.5 pts) have to be seperated then
traditionally some sort of tie-break mechanism is used to seperate the
players if it is to determine first place [2]. Other tied players are
left
tied in the competition results [2] because the fact is that swiss chess
has
not really had sufficient rounds to meaningfully seperate them. There
are
many of these tie breaks used in the chess world [3] but they all try to
estimate the player who has received the hardest draw and reward them.
This
is exactly the opposite effect of using a scoring system with more
possible
outcomes, eg 0-10, which will reward the player who received the softest
draw and allowed them to rack up points against the weaker players they
faced.
3. Current Team Scoring and Swiss Chess
Current practice for team scoring at DBM events is based on adding
together
the scores of the team members to get a team score. Hopefully it will be
obvious from the discussion in section 2 that this violates the pairing
assumptions of swiss chess in two ways:
1. It generates team scores which have a large range of possible
outcomes eg
for Granson 12-0, 11-1, 10-1, 9-3, etc. These effectively rate the team
against their precise set of opponents rather than against the whole
field
thus making correct pairing more difficult, increasing the probability
of
hard/soft draws among the teams and biasing the overall result towards
the
strong team with the softest draw.
2. By summing the team scores you are effectively averaging results and
thus
you are biasing further the distribution of results towards the mean. An
extreme way of looking at this is that all teams effectively get a draw
(average result) each round except for the teams that draw a team much
better/worse than them and these teams get a loss/win respectively. This
aggrivates the negative effects discussed in point 1 above.
The net result of these effects is a preponderance of noticably
hard/soft
draws at the competitions [4] and a huge advantage in the final placings
for
teams which get the opportunity to play against the weakest teams
present.
The winners, while undoubtedly strong teams, are also largely determined
by
the luck of the draw. Perhaps you should remind us again why this is so
-
that is direct the reader to the paragraphs in section 2.
4. Proposed Systems for Swiss Chess Team Competitions
In this section proposals are made to improve the reliability of the
results
produced at team competitions. Two proposals for changing the scoring
systems are discussed and also ways of restructuring the events while
retaining the current scoring methods (10-0? Or 310?) are mentioned.
4.1 New Scoring Methods
In order to satisfy swiss chess pairing requirements the current scoring
systems have to be changed to remove the current two problems discussed
in
section 3. Both of the suggestions here focus on extracting from the
individual game outcomes some sort of team outcome which is then used to
generate pairings and hence provides the primary scoring system for the
competition. There is still a huge amount of scope for different
proposals
on how exactly team scores should be structured (eg 1/.5/0 or 3/1/0),
how
individual games should be scored and what type of secondary (tie-break)
scoring system should be used.
4.1.1 Current Game Scoring Unchanged, Team Scoring Based on Sum
Under this proposal, games would be scored as per the current method and
then a team score is extracted from them. Take the results of all games
for
each team, sum them and compare to the other team's score. If:
Team 1's Total > Team 2's Total => Team 1 won, award them X points and
Team
2 lost, award them Z points.
Team 1's Total = Team 2's Total => Team 1 and Team 2 drew, award them Y
points.
Team 1's Total < Team 2's Total => Team 1 lost, award them Z points, and
Team 2 won, award them X points.
Using standard WDL scoring, X = 1, Y = 0.5, Z = 0. (Although other
schemes
such as 310 may be used by the organisers). This then becomes the team's
score for the round and is used to generate pairings. A secondary
scoring/tie-break mechanism would then be needed to seperate teams on
the
same score at the end of the competition. It is tempting to use the sum
of
the sum team members' scores for each round but of course this is biased
towards the team that got the softest draw and a better tie-break would
be
something like:
If team A beat team B during the competition, then team A places higher.
Otherwise take the sum of team A's opponents' scores and compare it with
the
sum of team B's opponents' scores, whichever is higher places higher.
This would be a simple and effective way of combating the current
problems.
It is dependant on how robust the individual game scoring mechanism is,
but
that is an issue for the organisers. It works well with swiss chess
because
it minimises the number of draws, which in turn minimises the number of
rounds required.
4.1.2 Holistic Approach, Raw Game Outcomes Give Team Score
A slightly more sophisticated approach might be to build a table of
individual game outcomes and map this onto the team outcome. An example
mapping for 4-person teams is given below:
Team A players get: Team A Result is: Team B Result is:
4 wins Win Loss
3 wins, 1 draw Win Loss
3 wins, 1 loss Win Loss
2 wins, 2 draws Win Loss
2 wins, 1 draw, 1 loss Win Loss
2 wins, 2 losses Draw Draw
1 win, 3 draws Draw Draw
1 win, 2 draws, 1 loss Draw Draw
4 draws Draw Draw
3 draws, 1 loss Draw Draw
1 win, 1 draw, 2 losses Loss Win
2 draws, 2 losses Loss Win
1 draw, 3 losses Loss Win
1 win, 3 losses Loss Win
4 losses Loss Win
Obviously many variations on this table are possible. Once the team
result
is decided, then score values such as the X, Y and Z in section 4.1.1
above
are allocated and used for pairing. A tie-break mechanism similar to
that
discussed in 4.1.1 would also have to be adopted. This is the preferred
approach of the author as it does not reward the ecentricities of
specific
game scoring mechanisms with the ability to push a team result from a
Draw
to a Win or Loss (as is the case with solution 4.1.1).
Avoiding the ability to use big wins on one table to more than offset
draws
or minor losses on another table seems to be an advantage in the same
way
that W-L is better than 10-0.
4.2 Other Approaches
Instead of changing the scoring system, there are at least two other
approaches that could be adopted. The easiest (although probably not
practical) approach would be to add more rounds to the competitions. How
many rounds are required? This is not easy to answer but we know that it
is
at least as many as WDL swiss chess, which would be log2(n) +
(log2(n))*1/3,
where n is the number of teams and probably less than a full round robin
which is (n-1) rounds. How many less will depend on the number of
possible
outcomes from each game, where more outcomes will require more rounds.
Unfortunately most organisers are stretched to fit in the number of
rounds
they play anyway so this is generally not a practical solution. A
corollary
of this approach would of course be to reduce the number of competing
teams
by at least 50%, it is assumed that this is also unpalitable to the
organisers.
The second approach, which was mentioned in BHGS list discussions [4],
would
be to drop the pairing by team aspect of the competitions, ie split each
team into the individual players who have their own scores and play
exculsively withing their own pools (ie all player 1's etc). This would
effectively split the competition into 3 or 4 seperate competitions with
individual winners. The team score would then be the sum of the final
scores
of each player in the team. This approach would certainly work but it
seems
to dilute the original concept of teams playing (and getting paired)
against
one another.
How does golfs Ryder Cup (US vs. Europe IIRC) handle this issue? Can
you
think of any other teamed sporting events to use as an analogy?
5. Conclusions
This paper has explained the relationship between scoring and pairings
in
swiss chess. It has also shown that getting pairing right is key to the
swiss chess mechanism functioning properly ie fairly determing a winner.
The
current practice of assigning a team score equal to the sum of
individual
player scores for team-based pairing purposes has been shown to interact
badly with swiss chess. Several proposals for alternative approaches to
team-based swiss chess competitions have been discussed. The author
recommends an approach based on extracting an overall team result from
the
individual game results which is then used for pairing and hence primary
scoring purposes. This approach maintains the team on team aspect of the
competition without requiring inpractical alterations to the competition
format and will increase the robustness of the final results generated.
Of
the two such approaches presented, the "Holistic Approach" is the
author's
preferred option.
References
[1] Olli Lounela, "Tournament Systems",
<http://kate.kttl.helsinki.fi/~olounela/tour.html>
[2] FIDE, "1992 FIDE swiss chess rules",
<http://members.ozemail.com.au/~tourney/pairingrules.html>
[3] Swiss Perfect, "Tie-breaks in swiss tournaments",
<http://members.ozemail.com.au/~tourney/tiebreak.htm>
[4] BHGS list discussion, 8th-14th Sept 2002,
<http://groups.yahoo.com/group/BHGS_UK_DBM_List/>